Two independent bugs made MCP terminal tools time out despite the command
finishing.
First, timeout_ms is declared integer in the tool schemas but was parsed
through GetStringArg, which only reads JSON strings. A numeric value fell
through to the 20s default, so callers asking for a longer wait were cut off
at 20s. exec_command, terminal_open, terminal_exec and remote_open now parse
timeout_ms through GetIntArg, which accepts both JSON numbers and digit
strings, keeping the 1..600000 range guard.
Second, FindSentinel treated the __MCP_DONE_<nonce>__ marker as a line start
only when preceded by \n. Commands that produce no output (ping > nul,
Start-Sleep, tar -czf) echo their command line ending in \r, so the marker
never matched and the wait ran to timeout even though the command had
completed. The check now also accepts \r.
Co-Authored-By: deepseek-v4-pro
Complete the bidirectional mutual exclusion between MCP remote control
and human Web viewing (§8.2). Direction 1 (remote_open rejected while a
human session holds the screen sub-connection) already existed; this adds
direction 2: a human Web viewer is now rejected while an MCP session owns
the device.
Add a m_McpTriggeredDevices marker (mirroring m_MfcTriggeredDevices) that
is set when an MCP session is created (BeginScreenCtrlOpen) and cleared at
every session-erasure site (CloseScreenCtrlSession, SweepIdleScreenCtrl,
OnScreenControlClosed, EndScreenCtrlAction), so the marker cannot go stale
and permanently block humans. HandleConnect checks IsMcpTriggered before
mutating client state or starting the remote desktop.
Co-Authored-By: deepseek-v4-pro
Add the remote_clipboard tool to the MCP remote control surface, writing
text to the remote host's clipboard via COMMAND_SCREEN_SET_CLIPBOARD
through the established screen sub-connection. Input UTF-8 is converted
to GBK (ToAnsi, code page 936) to match the client's CF_TEXT/ANSI
clipboard path, mirroring the existing CScreenSpyDlg::SendServerClipboard
packet format. Pasting remains a separate step (remote_keyboard Ctrl+V).
Known MVP limitation, documented in the tool description: non-GBK
characters (e.g. emoji) are replaced by '?' on the CF_TEXT path.
Co-Authored-By: deepseek-v4-pro
Add the remote_mouse tool to the MCP remote control surface, injecting
mouse events (move / down / up / click / right_click / middle_click /
drag / scroll) into an established remote_open session. Normalized 0..1
coordinates are mapped to physical pixels via the session's captured
resolution and clamped to screen bounds. Scroll follows the existing web
console's wheel sign convention (positive delta scrolls down) and is
vertical-only.
Extract BuildMouseMsg64 into WebService.h as a shared helper, reused by
both the web console's HandleMouse and the new MCP handler so the two
injection paths cannot drift.
Co-Authored-By: deepseek-v4-pro
Implement milestone M2b of the MCP remote-control design: inject keyboard
events through an existing screen sub-connection.
- Extract the MSG64 keyboard construction from WebService::HandleKey into a
shared inline BuildKeyMsg64 helper in WebService.h, and have HandleKey use
it (behaviour-preserving) so the Web and MCP paths cannot drift.
- Add BuildRemoteKeyboard with four actions: key_down / key_up / key_press
(key name -> VK via MapKeyNameToVk, plus CTRL/ALT/SHIFT/WIN modifiers) and
type (per-character VkKeyScanA mapping, ASCII only, newline/tab -> Enter/Tab).
The batch is sent as one [COMMAND_SCREEN_CONTROL][MSG64*N] packet over the
screen sub-connection under the existing Begin/EndScreenCtrlAction busy
discipline, then audited via WM_SHOWERRORMSG.
Non-ASCII text is rejected (-32602) and must go through remote_clipboard +
Ctrl+V (milestone M4), matching the design's clipboard path for CJK input.
Co-Authored-By: deepseek-v4-pro
Add the busy/closed flags to ScreenCtrlSession plus BeginScreenCtrlAction
/ EndScreenCtrlAction, mirroring the terminal's busy discipline so an
in-flight injection cannot race with session teardown (review finding #6).
While an injection is in flight (busy=true), OnScreenControlClosed marks the
session closed instead of erasing it (the injection thread, which still holds
subCtx, cleans up in EndScreenCtrlAction), SweepIdleScreenCtrl skips it, and
CloseScreenCtrlSession defers the erase. The subCtx is looked up under
m_ScreenCtrlMutex and the injection is sent outside the lock, matching the
established terminal pattern; the screen sub-connection's CONTEXT_OBJECT::Send2Client
already serializes internally via SendLock.
Co-Authored-By: deepseek-v4-pro
Add M1 of MCP remote control (docs/Mcp_RemoteControl_Design.md): the
remote_open / remote_close tools plus the ScreenCtrlSession state machine.
remote_open establishes a hidden screen sub-connection by reusing
WebService::StartRemoteDesktop (COMMAND_SCREEN_SPY -> CScreenSpyDlg ->
RegisterScreenContext), polls for the sub-connection plus its physical
resolution (TOKEN_BITMAPINFO -> NotifyResolutionChange -> GetScreenSize),
then records the session (single device, single session, reverse-mapped
subCtx for OfflineProc cleanup) and returns {session_id, screen_w,
screen_h}. remote_close validates session_id and tears down the
sub-connection idempotently. A McpRemoteControl settings checkbox (default
off, requires McpReadonly=0) gates the tools; every open/close is audited
via WM_SHOWERRORMSG.
Gating: multi-monitor hosts are rejected with -32008 (phase 1 supports only
single monitor, where Observe=main screen and Act=virtual desktop coincide);
the monitor count comes from the client heartbeat RES_RESOLUTION ("N:W*H").
Known limitations (deferred to the injection milestones): mutual exclusion
with human remote-desktop viewing is one-directional in M1 (a human who
joins during an MCP session can tear it down on disconnect), and subCtx is
not yet dereferenced so no liveness re-check is needed until
remote_mouse/remote_keyboard.
Co-Authored-By: deepseek-v4-pro
Add terminal_open / terminal_exec / terminal_close so an AI can hold one
shell session per Windows host and run a sequence of commands with cwd and
environment preserved, instead of the one-shot exec_command.
The persistent terminal is a separate full-command write capability gated by
McpTerminal (default off) plus McpReadonly=0, with no whitelist and full
audit. One device maps to one terminal session via the shared m_TermSessions
map; idle sessions are swept after 300s. terminal_exec rejects commands that
contain & or | (they corrupt the sentinel control-operator chain) as well as
control characters.
Also harden exec_command and terminal_exec against newline/CR injection, fix a
dangling subCtx after an abrupt shell disconnect, and refresh lastActiveAt on
command completion. Add en/zh-TW translations for the new UI strings and a
design document covering both exec_command and the persistent terminal.
Co-Authored-By: deepseek-v4-pro
Add a one-shot exec_command MCP tool that runs a command on a remote
Windows host and returns stdout plus exit code, reusing the Web terminal
link (main-connection COMMAND_SHELL, a shell sub-connection, and a
sentinel command line located via rfind to tolerate ConPTY echo).
The sentinel marker is embedded in the command line ConPTY echoes back,
so it is only treated as hit when it starts a line (preceded by a newline
or the buffer start); this keeps the echoed marker from being mistaken
for the real sentinel when the echo packet arrives before the output.
Execution is gated at Web remote-desktop sensitivity: a read-only mode
(McpReadonly, default on, hides the tool) and a command whitelist
(McpCmdWhitelist) with built-in read-only prefixes. Shell metacharacters
(& | < > ^) are rejected before whitelist matching, and each execution is
recorded in the server audit log.
Extend the MCP settings dialog with the read-only checkbox and a
multi-line whitelist box (commas and newlines both accepted, normalized
to a comma-separated list on save), and add English and Traditional
Chinese mappings for the new UI and audit-log strings.
Co-Authored-By: deepseek-v4-pro
Add a read-only list_registry MCP tool for querying a remote Windows
host's registry. An empty path returns the five root keys; a path such as
HKEY_LOCAL_MACHINE plus subkeys returns that key's immediate subkeys and
values (name, type, and formatted data).
Client side: enumerate with KEY_READ instead of KEY_ALL_ACCESS and map the
full value-type set (REG_SZ, REG_DWORD, REG_BINARY, REG_EXPAND_SZ,
REG_MULTI_SZ, REG_QWORD, REG_NONE) so unknown types are no longer
mis-reported as REG_SZ. The registry manager now always sends both the
TOKEN_REG_PATH and TOKEN_REG_KEY packets (an empty packet when a part is
empty), making the two-packet reply deterministic for the server.
Server side: add the three-phase flow (COMMAND_REGEDIT, COMMAND_REG_FIND,
then PATH plus KEY) with per-host pending state, parse the fixed-width
wire format with bounds checks, format value data for JSON, and guard
registry paths against exceeding MAX_PATH to protect the client stack.
Co-Authored-By: deepseek-v4-pro
Expose the server main-window audit log (host online/offline, operation
results, alerts) as a read-only, no-arg, in-memory MCP tool. The log lives
in m_CList_Message, a CListCtrl that is only safe to touch on the UI
thread, so add a thread-safe mirror m_MessageLog (guarded by m_cs) and keep
it in sync at the four item-mutation sites: ShowMessage, OnShowErrMessage,
OnMsglogDelete and OnMsglogClear. The tool reads the mirror directly under
m_cs and returns up to MAX_MESSAGE_COUNT (1000) entries, newest first.
Co-Authored-By: deepseek-v4-pro
Add the two P2d read-only MCP tools, both one-shot sub-links (mode A') on
the existing single-flight pending registry.
list_services sends COMMAND_SERVICES; the client's CServicesManager emits
TOKEN_SERVERLIST on sub-link creation, parsed as 5 null-terminated fields per
record (display_name/service_name/binary_path/status/start_type) with
zero-padding termination. Services are Windows-only, so LNX/MAC hosts get
-32005 up front instead of a 20s timeout.
get_client_log sends COMMAND_QUERY_LOG; the client's CClientLogManager dumps
its full in-memory Logger ring buffer once, then pushes deltas every 3s. The
MCP path takes the first (full) TOKEN_REPORT_LOG and cancels the sub-link so
later deltas stop, while the MFC log dialog keeps receiving deltas on the
non-pending branch. Log text is client ANSI on Windows, decoded by clientType
like process/file names. MessageHandle intercepts both with
IsPending -> TakeMainResponse + CancelIO. Sync the design doc (P2d section +
verification notes; Linux get_client_log timeout is a client-version gap).
Co-Authored-By: deepseek-v4-pro
The get_screenshot tool inherited the MFC thumbnail-preview profile, capping
the frame at 1024px wide (1280 on 4K source), which is too small for AI
vision/OCR of dense UI. Add an optional max_width argument that overrides the
width (clamped to the client's 64..1920 limit) while keeping the RTT-adaptive
jpegQuality. Omitted or 0 keeps the existing thumbnail profile; 1920 yields
near-native resolution (full 1080p on a 1080p source, 1920 wide on 4K). Sync
the design doc.
Co-Authored-By: deepseek-v4-pro
Add the two P2c MCP tools on top of the P2b protocol. list_files lists
drives (COMMAND_LIST_DRIVE -> TOKEN_DRIVE_LIST) or a directory (follow-up
COMMAND_LIST_FILES -> TOKEN_FILE_LIST on the same one-shot sub-link);
get_screenshot reuses the screen-preview RPC with a per-host reqId
correlation so stale or MFC-preview responses fall through to the MFC
path untouched. Both share the P2b single-flight pending registry.
Fix file/process name encoding: the client reads process names, paths and
file names via the ANSI (A) APIs, so they are GBK on Windows regardless of
the CLIENT_CAP_UTF8 capability bit (which governs window titles only).
Decode by clientType (LNX/MAC = UTF-8, Windows = 936) rather than
GetClientEncoding, and convert the list_files path to the client ANSI code
page before sending. Sync Mcp_Phase2_Design.md with the P2c design and the
encoding correction.
Co-Authored-By: deepseek-v4-pro
Add three read-only P2b MCP tools over the existing protocol. list_processes and list_windows trigger the client via COMMAND_SYSTEM / COMMAND_WSLIST on the main connection and receive TOKEN_PSLIST / TOKEN_WSLIST on a one-shot sub-link; get_activity_history uses the main-connection RPC COMMAND_QUERY_ACTIVITY -> TOKEN_REPORT_ACTIVITY. A per-host single-flight pending registry (m_Pending) with a 20s timeout correlates each response to its request and rejects a concurrent request for the same host with -32003. Parsers stop on the first empty record to ignore the client's LocalSize trailing zero padding, and window titles are decoded per the client UTF-8 capability bit. MessageHandle only adds if-guarded branches, so the MFC dialogs are untouched. Sync Mcp_Phase2_Design.md with the mode A'/A architecture, the verification notes, and the registry-backed config location.
Co-Authored-By: deepseek-v4-pro
Add two pure in-memory MCP tools over the online host list. search_hosts filters by name/remark, IP, group and OS (all optional, AND-combined, ASCII case-insensitive substring match); get_host_detail returns one online host by id and answers -32602 for a missing/non-numeric id and -32002 for an unknown or offline id. Sync Mcp_Phase2_Design.md with the second-review corrections and the P2a implementation.
Co-Authored-By: deepseek-v4-pro
Add an optional MCP server (JSON-RPC 2.0 over Streamable HTTP) exposing
an online-host listing tool, protected by a Bearer token. Disabled by
default; configured via a new "Extensions > MCP Settings" dialog.
- McpServer: httplib + JSON-RPC 2.0 dispatch (initialize/ping/tools/list/tools/call)
- McpSettingsDlg: runtime-created dialog for enable/port/bind/token
- HostJson: extract single-host JSON serialization shared with WebService
- FRP: expose MCP port (union with listening/Web ports) when bound to 0.0.0.0
- i18n: en_US / zh_TW translations
Co-Authored-By: deepseek-v4-pro
